Stochastic Deep Learning for Compressive-sensing Radar

Machine deep learning (DL) is very good in fitting nonlinear functions and learning data representations. It is still unclear why DL works well because the statistical properties have not been fully understood yet. Stochastic deep learning (SDL) is proposed which bridges the strong numerical capability of DL with the probabilistic inferences. Since signal models in radar are intrinsically nonlinear with respect to unknown target parameters (range, Doppler or angles), and moreover, radar processing is intrinsically stochastic, SDL can enhance radar, especially, radar with compressive sensing (CS) whose data are more complicated due to the signal compression. The potential of SDL applicable to CS radar is presented.